tls@rek.tjls.com wrote:

> I tried this several months ago.  It seemed to cause problems.  But
> I rewrote the code instead of using it directly from the Broadcom driver
> (I don't think OpenBSD had it at all then) so maybe I got it wrong.

Hmm. Was it shared interrupt case?

> I believe Jonathan also said that checking the BGE_PCI_PCISTAT
> register might be quite expensive.

The comment says checking BGE_MISC_LOCAL_CTL is expensive.
Is it also true for BGE_PCI_PCISTATE?

If interrupts are for bge(4) itself, bge_status is updated
in most case so BGE_PCI_PCISTATE is rarely checked.

If interrupts are not for bge(4), the register is checked
always, but is it more expensive than calling all unneeded
functions (and read/write against some registers) in handlers?
---
Izumi Tsutsui